Training Jobs, checkpoints, PVC storage, and automated recovery
Practice self-check. Answers stay in this page so you can grade yourself. This is not a scored exam.
Twelve scenario items on Kubernetes Jobs and storage plus AWS automated recovery for RTO/RPO. Companion lab: Hedronite trainjob (checkpoint resume, backoffLimit, honest drift).
Lab download.Starter is the tree you edit. Worked answer is a separate file; open it after the starter passes. Rust 1.85 or newer, tar, and python3 for walkthrough.sh. One online cargo vendor vendor. The zip has no vendored crates. All lab requirements.
Q1 · CKA workloads and scheduling, Job backoffLimit
A platform team runs a training Job whose manifest omits .spec.backoffLimit and also sets no activeDeadlineSeconds. After several Pod failures, an engineer asks how many retries the Job controller allows before it marks the Job as failed. Which statement matches the Kubernetes documentation?
Answer: B. The Job documentation states that .spec.backoffLimit defaults to 6, after which the Job is considered failed. Source: Jobs, section Pod backoff failure policy: - https://kubernetes.io/docs/concepts/workloads/controllers/job/#pod-backoff-failure-policy
Q2 · CKA storage, PersistentVolumeClaim
A team needs durable storage for training checkpoints. The storage must be requested by a user through a namespaced API object and satisfied from cluster PersistentVolume resources, rather than from node CPU or memory. Which action meets this requirement?
Answer: D. The documentation defines a PersistentVolumeClaim as a request for storage by a user, and says PVCs consume PV resources the way Pods consume node resources. Source: Persistent Volumes, section Introduction: - https://kubernetes.io/docs/concepts/storage/persistent-volumes/#introduction
Q3 · DOP-C02 automated recovery (3.3), EBS snapshot recovery point
A training node is an EC2 instance whose EBS data volume holds model checkpoints. The instance AMI was created before any checkpoints were written. The team needs a point-in-time backup so that, if the disk is later corrupted, they can recreate the volume as it was at backup time. Which approach meets this need?
Answer: A. An EBS snapshot is a point-in-time backup of a volume, and you can create a new volume from a snapshot to restore that state. Source: Create Amazon EBS snapshots: - https://docs.aws.amazon.com/ebs/latest/userguide/ebs-creating-snapshot.html
Q4 · CKA workloads and scheduling, Job Pod recreation
A training Job has a completion count of 1 and a backoffLimit of 6. Its only Pod fails because the node it runs on is rebooted, and the Job has not yet reached the backoffLimit. What does the Job controller do?
Answer: C. The Job documentation says that when a whole Pod fails, for example because its node is rebooted, the Job controller starts a new Pod, counting the failure toward the backoff limit. Source: Jobs, section Handling Pod and container failures: - https://kubernetes.io/docs/concepts/workloads/controllers/job/#handling-pod-and-container-failures
Q5 · CKA storage, PVC vs emptyDir
A training Job writes checkpoints of about 200 MiB. When the Job creates a replacement Pod, the old Pod is removed from its node, and the new Pod must still see the checkpoint data. The team is currently using an emptyDir volume. What should they do?
Answer: A. A PersistentVolume has a lifecycle independent of any individual Pod, whereas emptyDir data is deleted permanently when the Pod is removed from the node. Source: Persistent Volumes, section Introduction: - https://kubernetes.io/docs/concepts/storage/persistent-volumes/#introduction
Q6 · DOP-C02 automated recovery (3.3), RPO and backup frequency
Training artifacts live on EBS volumes, and the organization's recovery point objective (RPO) is at most one hour of data loss. An Amazon Data Lifecycle Manager policy currently creates one snapshot every 24 hours. The data must stay on EBS, and only the backup configuration may change. What change moves the design toward the RPO?
Answer: B. RPO is bounded by the gap between recovery points, and a Data Lifecycle Manager schedule can run as often as every hour, which shrinks that gap. Source: Amazon Data Lifecycle Manager: - https://docs.aws.amazon.com/ebs/latest/userguide/snapshot-lifecycle.html
Q7 · CKA workloads and scheduling, Job restartPolicy
An engineer drafts a Job manifest for a batch trainer and sets restartPolicy: Always in the Pod template. The API server rejects the manifest. Which statement describes what the Kubernetes documentation allows for a Job's Pod template?
Answer: D. The Job Pod template documentation states that only a restartPolicy equal to Never or OnFailure is allowed. Source: Jobs, section Pod template: - https://kubernetes.io/docs/concepts/workloads/controllers/job/#pod-template
Q8 · DOP-C02 automated recovery (3.3), restore testing for RTO
A team has AWS Backup recovery points for the EBS volumes that hold model checkpoints. Leadership wants evidence, before any incident, that restores will complete within the recovery time objective (RTO). The evidence must be gathered without affecting production volumes. What should the team do?
Answer: C. AWS Backup restore testing performs automated test restores of recovery points so the team can see whether restore times meet the organizational RTO. Source: AWS Backup, Restore testing: - https://docs.aws.amazon.com/aws-backup/latest/devguide/restore-testing.html
Q9 · CKA storage, claims as volumes
A PersistentVolumeClaim named train-ckpt is already bound in the training namespace. A training Pod in the same namespace must use that durable storage for checkpoints. How should the Pod spec consume it?
Answer: B. Pods access storage by using the claim as a volume, with the claim in the same namespace referenced through persistentVolumeClaim.claimName. Source: Persistent Volumes, section Claims As Volumes: - https://kubernetes.io/docs/concepts/storage/persistent-volumes/#claims-as-volumes
Q10 · CKA workloads and scheduling, Job for batch work
A one-off training task must run to completion and then stop. If a Pod fails, it must be retried, and the work must not be lost to a node reboot without a replacement Pod. Which Kubernetes object fits this workload?
Answer: A. A Job creates Pods that run to completion, retries failures, and replaces Pods lost to node problems until the completion criteria are met. Source: Jobs, section Alternatives: - https://kubernetes.io/docs/concepts/workloads/controllers/job/#alternatives
Q11 · DOP-C02 automated recovery (3.3), failover and restore tests
A platform team runs Multi-AZ training infrastructure with AWS Backup recovery points. Before declaring disaster recovery ready, they must validate that recovery meets the stated RTO and RPO targets. An auditor will not accept written procedures alone as proof. What should the team do?
Answer: D. Well-Architected guidance and AWS Backup restore testing both call for exercising failover and restores so that measured recovery time and data loss can be compared with the targets. Source: AWS Well-Architected Reliability Pillar, REL13-BP03 Test disaster recovery implementation to validate the implementation: - https://docs.aws.amazon.com/wellarchitected/latest/reliability-pillar/rel_planning_for_recovery_dr_tested.html
Q12 · CKA storage, PV reclaim Retain
After a successful training run, the team deletes the PersistentVolumeClaim. The operations team must keep the underlying PersistentVolume and its data so an administrator can reclaim it manually later. They want to use a reclaim policy that is not deprecated. Which setting meets this need?
Answer: C. With the Retain reclaim policy, the PV and its data remain after the PVC is deleted, and the volume is considered released until an administrator reclaims it manually. Source: Persistent Volumes, section Reclaiming (Retain): - https://kubernetes.io/docs/concepts/storage/persistent-volumes/#retain